CODEWITHSUNDEEP
pythonmachine-learningpytorchneural-network
smaragda ben
smaragda ben

Reputation: 21

TypeError: __init__() got an unexpected keyword argument 'device_type' in torch.cuda.autocast

with autocast(device_type='cuda', dtype=torch.float16):
    a=0.00000000000000001
    print(a)

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: __init__() got an unexpected keyword argument 'device_type'

I got this error in another tool I was using so I wanted to reproduce it and I run the above dummy code and I got it again. Anyone knows if it's a pytorch, cudatoolkit problem or something else? These are the relative packages I have in my anaconda environment:

pytorch 1.9.0 py3.8_cuda11.1_cudnn8.0.5_0 pytorch python 3.8.15 h7a1cb2a_2 cudatoolkit 11.1.74 h6bb024c_0 nvidia

And this is the error I initially came across in the training loop of a classification model I was using.

with autocast(dtype=self.precision):
TypeError: __init()__ got an unexpected keyword argument 'dtype'

Upvotes: 2

Views: 5336

Answers (1)

Aman Maghan
Aman Maghan

Reputation: 716

If you're intending to use autocast module from pytorch, use it as following-

torch.autocast(device_type='cuda', dtype=torch.float16)

Upvotes: 1

Related Questions